Test generation across multiple layers

Abstract

Complex software systems frequently come in many layers, each realized in a different programming language. This is a challenge for test generation, as the semantics of each layer have to be determined and integrated. An automatic test generator for Java, for instance, is typically unable to deal with the internals of lower level code (such as C-code), which results in lower coverage and fewer test cases of interest. In this paper, we sketch a novel approach to help search-based test generators for Java to achieve better coverage of underlying native code layers. The key idea is to apply test generation to the native layer first, and then to use the inputs to the native test cases as targets for search-based testing of the higher Java layers. We demonstrate our approach on a case study combining KLEE and EVOSUITE.